[Debtags-devel] Committed 'debtags ssearch'

[Enrico Zini]

Hello,

I did an interesting commit to the debtags commandline tool: a new
command called 'ssearch' (smart search).  It works like this: ssearch
takes arguments.  If they have a + in the front, they are wanted tags.
If they have a '-', unwanted tags.  Else, they're keywords.

More than that, it gives on output also the packages with a similar
tagset to the most common tags in the output, and it shows what these
'most common tags' were.  That way you can rerun the search with
refining based on the tags you see in the output.

It's a start to new generation search ideas.

The test case I had for it was to find 'gimp' searching for 'image
editor' without using tags that I don't see in the output.
It went like this:

  debtags ssearch image editor
  [no gimp on the results, but various image editors mainly for kde.
   Most common tags are remarkably accurate, though]

  debtags ssearch image editor -suite::kde
  [still no gimp on the results, but I consider being heavier on the
   tags I see]

  debtags ssearch image editor -suite::kde +use::editing +works-with::image::raster
  [no output at all: I realize the keyword search always blocks gimp for
   coming out.  But the tags I have now are good enough]

  debtags ssearch -suite::kde +use::editing +works-with::image::raster
  [there!]

Now, I'd like to make it even smarter, but for today I did enough.  I
guess there's some refining to do on the 'related packages' part.

BTW, I've implemented a scored tagset distance algorithm.  Now one could
give different scores to facets (or tags) to say how relevant they are
in their judgement.  This allows us to assign 0.1 score to
implemented-in, for example, to make it less relevant than, say, use::
and works-with::.


Ciao,

Enrico

--
GPG key: 1024D/797EBFAB 2000-12-05 Enrico Zini <enrico at debian.org>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
Url : http://lists.alioth.debian.org/pipermail/debtags-devel/attachments/20051022/5c3ceaf6/attachment.pgp